Nvidia has commenced pre-orders for its latest China-specific artificial intelligence (AI) chip, the H20, through distributors who are pricing it on par with a rival product from Huawei, according to sources familiar with the matter.
The H20, touted as the most powerful of the three Nvidia chips developed for the Chinese market, comes amidst expanded U.S. bans on high-end chip exports aimed at hampering China’s AI and sophisticated computer development for military purposes. While the H20 is expected to deliver less computing power than Nvidia’s flagship H100 AI chip and the H800, which was banned in October, it is positioned to address the growing demand for AI capabilities within China.
However, sources indicate that the H20 may be less powerful than Huawei’s Ascend 910B in certain key areas. Specifications suggest that the H20 lags behind the 910B in FP32 performance, a critical metric for chip processing speed, but has an advantage in interconnect speed, crucial for data transfer between chips.
Despite this, Nvidia has been pricing the H20 aggressively, with pre-orders ranging from $12,000 to $15,000 per card. Distributors in China have already started advertising the chips with markups, some as high as 110,000 yuan ($15,320), comparable to the price of Huawei’s 910B, which is selling for around 120,000 yuan.
Additionally, Nvidia is offering pre-configured H20 servers with 8 AI chips at 1.4 million yuan, positioning them as a more cost-effective solution compared to servers using 8 H800 chips, which were sold at around 2 million yuan a year ago.
The rollout of the H20 is seen as crucial for Nvidia, which previously dominated China’s AI chip market with over 90% share but now faces increased competition from domestic rivals, especially Huawei. The Chinese tech giant’s 910B chip is considered highly competitive and has gained popularity amid concerns about restricted access to Nvidia’s products due to U.S. sanctions.
Nvidia’s CEO, Jensen Huang, recently visited the company’s offices in Shenzhen, Shanghai, and Beijing, underscoring the importance of the Chinese market. The H20 is expected to enter mass production in the second quarter of 2024 after a delay in its original November launch, attributed to integration issues with server manufacturers.
In addition to the H20, Nvidia plans to release two other China-specific chips, the L20 and the L2, although details about their rollout status remain undisclosed. None of these chips are currently listed on Nvidia’s website.
As Nvidia navigates the competitive landscape in China’s AI chip market, the success of the H20 will likely play a significant role in shaping the company’s position in this critical sector.
(Source: Reuters | TipRanks | Seeking Alpha | Asia Business Outlook)
